Trust
What happens to your manuscript
Unpublished work is sensitive. This page explains, in plain terms, where your files go, who processes them, what we keep, and what we never do with them.
Our commitments
Your research is never used to train AI
These are restated from our privacy policy, which is the binding document.
Never used to train AI
Your research is never used to train AI models, ours or anyone else's.
Business APIs only
We access OpenAI, Anthropic (Claude) and Google (Gemini) only through their business APIs, whose terms exclude customer data from model training. Providers may keep requests for a short period (typically up to 30 days) solely for abuse and safety monitoring, then delete them.
Yours, not ours
Your ideas, drafts and results stay yours. We never share or publish them, and we never sell your personal data.
Encrypted
Data is encrypted in transit and at rest, on Firebase / Firestore cloud infrastructure.
Step by step
The life of an uploaded file
- Upload. You upload a PDF, DOCX or TXT file (up to 10 MB) over an encrypted connection.
- Text extraction. Our server extracts the plain text from the file. The file passes through the server’s temporary disk while this happens; the original file is not saved to our database or to any file storage.
- AI analysis. The extracted text, or the relevant part of it, is sent to the AI providers that tool uses (see the table below) through their business APIs. They do not train on it.
- Verification. Where a tool checks references, only citation details (titles, authors, years, DOIs) or short search queries are sent to the scholarly databases — not your manuscript.
- Results saved to your account. Reports are saved so you can come back to them. What is saved differs by tool, and we list it exactly below.
Per tool
Who processes your text, and what we keep
Every tool also writes a small usage record (task, cost, time) for billing and your credit history.
| Tool | AI providers | What is stored |
|---|---|---|
| Verify Citations | OpenAI (extracts the reference list) | A report with the file name, the extracted citations and their verification results. It does not contain your manuscript text. The report has a long random ID and anyone with its link can open it. |
| Suggest Citations | OpenAI, Anthropic, Google | A usage record (task, cost, time). The manuscript text is not stored. |
| Draft Feedback | OpenAI | The file name and the feedback, in your account history. |
| AI Peer Review | OpenAI, Anthropic, Google | The reviews and scores, plus the first 10,000 characters of the paper text so a revised version can be compared with the earlier one. |
| Deep Audit | OpenAI, Anthropic, Google | The file name, your settings and the audit report. The manuscript text is not stored. |
| Submission Readiness Check | OpenAI, Anthropic, Google | The extracted manuscript text and any journal guidelines you paste, kept in a server-only record so each stage of the check can run, plus the results. The report is private unless you turn on sharing. |
Sub-processors
The services we rely on
We share data with these services only to run the product. No data is shared with third parties for marketing, and we use no third-party advertising trackers.
Google Cloud / Firebase
Account sign-in, database (Firestore) and product analytics (Firebase Analytics).
OpenAI, Anthropic, Google (Gemini)
AI processing through business APIs, as described above.
Render
Hosts the web application. Uploaded files are processed on these servers.
Stripe
Payments. Card details go to Stripe; we do not store them.
Brevo
Transactional email, such as account verification and service updates.
Sentry
Error monitoring, so we can find and fix crashes.
Verification
How references are checked
Language models can invent references. So we check every citation against published records. On the review tools, findings are also cross-checked across multiple models.
CrossRef
DOI lookup and title search — the authoritative source when a DOI is present.
Semantic Scholar
Fallback title search when CrossRef has no match.
OpenAlex
Second fallback title search.
Retraction checks
Every resolved DOI is checked in CrossRef for retraction, withdrawal and expression-of-concern notices (CrossRef now carries the Retraction Watch data).
Each citation ends up verified, suspicious (found, but a detail such as the year or first author doesn’t match), unverifiable (no matching paper found) or retracted. “Unverifiable” does not mean fake — books, reports and very new papers are often missing from these databases. Read the technical overview
Your rights, including deletion
- Access your personal data at any time.
- Request correction or deletion of your data.
- Export your research content.
- Withdraw consent for data processing.
Our privacy policy commits us to the GDPR. To delete your account and data, email privacy@researchbuddyai.com.
Questions about your data?
If your institution needs more detail before you upload, ask us. We’ll answer specifically.